Aquasi-naturalexperimentontheflypapere ect:the2008 local fiscal reform in Italy

Transcription

1 Aquasi-naturalexperimentontheflypapere ect:the2008 local fiscal reform in Italy Massimiliano Ferraresi, a Umberto Galmarini, b Leonzio Rizzo c and Alberto Zanardi d a University of Ferrara Ferrara, massimiliano.ferraresi@unife.it b University of Insubria Como and IEB Barcelona, umberto.galmarini@uninsubria.it c University of Ferrara Ferrara and IEB Barcelona, leonzio.rizzo@unife.it d University of Bologna Bologna and Econpubblica Bocconi Milano, alberto.zanardi@unibo.it April 2015 Abstract We investigate the impact on expenditure of tax on principal dwellings before 2008 and the impact on expenditure of the grant which, after 2008, compensated for the abolition of the tax on principal dwellings. We setup a theoretical model in which the introduction of a political bias against taxation gives rise to the flypaper e ect. If the public good is very important with respect to private consumption then an increase in the municipal size implies a decrease in the extent of the flypaper e ect; the opposite happens if the public good is not important with respect to private consumption.we then test the hypotheses coming from the model by using data on Italian municipalities, focusing on two groups of expenditure: the principal expenditure, which are those essential to guarantee the minimum standard daily life of a municipality and the rest, defined as residual expenditure. We find that the flypaper e ect holds for both kinds of expenditure, but decreases with respect to population in the case of principal expenditure and increases with respect to population in the case of residual expenditure. Keywords: Flypaper, transfers, federal budget, ICI, fiscal reform. 1

2 1 Introduction In 2006, just before the end of the election campaign, Berlusconi, the right-wing candidate for Prime Minister, said If you vote for us again, we will abolish property tax for your primary residence. There is evidence in Italy (Bordignon and Piazza, 2010) that this tax is a salient political issue at local level, 1 in fact this claim bought homeowners votes for the right-wing candidate; however, on the other hand, this striking proposal put local governments in a state of uncertainty, since property tax is a very important source of funding for municipalities. Two years later, in 2008, the central government in Italy totally exempted citizens from the payment of the property tax (ICI) levied on principal dwellings, thus leading to a significant decrease in the availability of municipalities own resources, which were replaced by a compensating transfer from the central government. Such a change in fiscal policy allows us to investigate the impact of the municipal revenue linked to principal dwellings (either raised by municipalities before 2008, or funded through the central transfers after 2008) on local expenditure. Federal grants distributed to members of a federation should only alter income levels and a ect state expenditure in the same way lump-sum grants to individual community members would (Bradford and Oates, 1971). However, empirical works in the field do not support this theory and one of the most accredited alternative explanations is the flypaper e ect. Grants stimulate government expenditures more than transfers to individuals for the same amount of money (Gramlich, 1977). Hence, a proportion of federal money remains in the public sector rather than of being distributed among citizens. In seminal empirical works, Henderson (1968) and Gramlich (1969) found that an extra dollar of personal income increased government spending from $0.02 to $0.05 but an equivalent extra dollar of grants increased government spending by $0.30: this larger e ect of lump-sum aid on government spending was then called flypaper e ect following Arthur Okun s observation that money seems to stick where it hits. Starting from these findings, much literature has developed documenting and seeking to explain the flypaper e ect. 2 According to Inman (2009), the flypaper e ect can arise for four reasons. 1 According to Corriere della Sera the most popular Italian newspaper this tax is considered as the most hated tax by Italian taxpayers (Corriere della Sera, May 22, 2007). 2 For a comprehensive analysis see, e.g., Hines and Thaler (1995), Gamkhar and Shaw (2007) and Inman (2009). The 2

3 first one concerns the data: researchers might confuse matching grants with lump-sum grants or may be particularly sensitive to some kind of transfers as Wycko (1991) finds for capital expenditures. The second explanation relies on a possible econometric mis-specification as empirical studies on the flypaper e ect often omit important unobserved input variables (Becker E. 1996; Megdal S. B., 1987; Zampelli E. 1986). A new interesting explanation, related to this second reason, comes from the idea that federal transfers can be endogenous in a regression of the local expenditure (Knight, 2002): a positive correlation between constituent preferences for public goods and intergovernmental grants biases upwards the coe cient relating federal transfer to local expenditure. The third explanation is based on the voter ignorance hypothesis. The representative voter does not know the level of grants received by the local government which it cannot then include in its private budget constraint, or, as stated by Hines and Thaler (1995) the representative voter is aware of the aid received by the local government but distinguishes between public budget, which is the responsibility of government o cials, and a private budget, which is the citizen s responsibility, meaning that only part of the grant is included in the private budget. Finally, according to the fourth explanation, the flypaper e ect is a consequence of an inability of citizens to write complete political contracts with their elected o cials because they have imperfect information about intergovernmental grants and budget-maximizing bureaucrats who use hidden information to expand their budget (Wycko, 1988). Besides these explanations, part of the literature points out that the flypaper e ect can arise where subnational governments use distortionary taxes to fund their expenditure (Hamilton, 1986; Becker and Mulligan, 2003; Voleden, 2007) and, at the same time, receive federal grants, which are very di cult, for the citizens, to relate to the federal taxes they pay, hence, they are perceived as lump sum grants. These grants in addition to the distortionary taxes intended to finance the public good lead not only to the classical income e ect, but also to a price e ect, decreasing the marginal cost of public funds (Dahlby, 2011). There is a large amount of literature testing the flypaper e ect. In particular, Winer (1983), using data on Canadian provinces for the period , shows that the e ect of grants on provincial spending for poor provinces is about two times larger than that for the rich provinces. Blanco (2006) finds that the flypaper e ect in Brazil is more marked in municipalities with a low level of population density. Buettner and Wildasin (2006) use a panel dataset of 1270 U.S. municipalities over the period

4 1997 finding that a permanent one dollar per capita increase in grants leads to a 28.7 cent increase in spending and, interestingly, this e ect is more pronounced for large US cities compared to small ones. Kalb (2010) uses data on German municipalities and shows that an increase in the amount of grants received by the local government implies not only an increase in expenditure, but also a loss in productive e ciency. In relation to the Italian case, Levaggi and Zanola (2003), using data at regional level from 1989 to 1993, find evidence of the flypaper e ect for health expenditure. Revelli (2013) shows how excess sensitivity of local public spending to grants arises in the presence of tax limitations. By using data for the Italian provinces over the years 2000 to 2007 he finds that the response of local spending to grants is significantly higher for fully constrained provinces than for provinces that can handle at least one tax instrument. Finally, Gennari and Messina (2014) test the presence of flypaper also investigating the role played by some political factors like the electoral cycle or the political strength of the local cabinet, by using data on Italian municipalities from 1999 to 2003 and, find a strong flypaper e ect but that is not a ected by political factors. In this work we exploit a sort of quasi-natural experiment since an exogenous change in fiscal policy allows the expenditure of municipalities to be compared based on two di erent financing systems: one based on own revenue (pre-2008) and the other based on vertical transfers (post-2008). The work is structured as follows. Section 2 presents the theoretical model. Section 3 discusses the fiscal policy reform and provides some institutional information on Italian financing systems as well as a description of the data. Some preliminary evidence is illustrated in Section 4. Our empirical strategy and results are in Section 5. Section 6 is the conclusion. 2 The theoretical model In this Section we use a neoclassical model similar to the one in Dahlby (2011). However, in Dahlby s model the flypaper e ect arises due to the fact that a benevolent local government uses a local tax which is distortionary. In the model that follows, taxation is instead non-distortionary, since we focus on taxes on principal dwellings, but we introduce a political bias against local taxation that gives rise to the flypaper 4

5 e ect. 3 Consider a municipality. The welfare of the municipality is represented by the quasi-concave utility function u(c, G), where c is per capita private consumption and G is the public good. The municipal government finances the public good with a tax on principal dwellings and with a transfer from the central government. The per capita local tax base, b, is exogenously given, and the tax is proportional, at rate. The budget constraint of the private sector is c = y b, where y is the per capita income of the municipality, exogenously given. The budget constraint of the municipality is G =( b+ t)n, where t is the per capita grant from the central government, in lump sum form, and N is the size (population) of the municipality. The local government s objective function is V = u(c, G) l( b) where l( b) is a loss function that captures citizens aversion to taxation, strictly convex in tax revenues. This function captures in a reduced form the bias that citizens have when evaluating fiscal policies: they overvalue the costs of taxation while they undervalue the benefits of the public good. The policy maker maximizes her political support by maximizing true social welfare u(.) while minimizing the unpopularity stemming from taxation. To illustrate, consider the following quadratic specification u(c, G) = c + (1 )c 2 l( b)= 2 ( b) 2 G 2 G 3 This feature of our model has some evidence in Italy, where municipalities, when increasing tax, usually prefer to increase the surtax on national income tax than local property tax on dwellings, since the former, even if it is formally a local tax, is perceived as a national one hence not related to local policy maker behavior (Bordignon and Piazza, 2010). 5

6 where >0, 0 < <1 are parameters characterizing the preferences for the private and the public good, and 0 is a parameter capturing the degree of aversion to taxation. From the first order condition with respect to the tax rate we get: hence: (y, t)b = (1 )y + (N 1) N2 t 1+ + (N 2 ; (1) 1) G (y, t) = [(1 )y + (N 1) + (1 + )t] N 1+ + (N 2. (2) 1) In the absence of transfers, i.e., if t = 0, the provided public good (2) is N times the optimal raised per capita revenue, (y, 0)b. In fact, it is easy to see that in this case: (y, 0)b = (1 )y + (N 1) 1+ + (N 2 1), (3) and, the total revenue, which coincides with the provided public good when t = 0, is equal to (y, 0) = [(1 )y + (N 1)] N 1+ + (N 2. 1) If we introduce a transfer t>0, the optimal raised revenue (1) is lower than the optimal raised revenue when no grant holds (3) because the grant (t >0) increases the available total revenue not a ecting the local political cost of taxation and therefore the local policy maker needs less taxes to finance any given public good. Moreover, since the policy maker knows that increasing the provision of the public good through the increase in transfer does not a ect private consumption, she will choose a higher level of public good than in the case when there was no grant. In fact, using (3), we can re-write (2), as follows: G (y, t) = (y, 0) + (1 + )tn 1+ + (N 2 1), which states that the provided public good when a transfer holds is higher than the public good provided when a transfer does not hold ( (y, 0)). We are interested in comparing the change in G (y, t), when a change in local tax revenue is exogenously induced by, for example, an increase in y and comparing it, with the case when an increase in t is introduced. In the absence of political aversion to 6

7 taxation (i.e., = 0), since taxes are non-distortionary (i.e., tax bases are exogenous), we do not observe the flypaper e ect, that is: Instead, = @y =0 (1 )N 1+ (N 2 1). >0, which means that there are political costs in raising local taxation, we have the flypaper e = (1 + )N 1+ + (N 2 1) > (1 )N 1+ + (N 2 Transforming one unit of income into public good is more expensive (because citizens must be locally taxed), than transforming one unit of transfers in public good, which does not have any political cost for the local policy maker. flypaper e ect is more marked the larger and: = (1 )(1 + N2 ) [1 + + (N 2 = (1 + )(1 + N2 ) [1 + + (N 2 Note in fact = (1 + N2 ) [1 + + (N 2 1)] 2. (4) @y@n < 0 if and only if > ; in this case an increase 1+N 2 in the municipality size decreases the size of the flypaper e ect: if the public good is very important (i.e., > 1+ 1+N 2 ) and so a significant proportion of the private income (through taxation) has already been allocated to finance it, a further increase in population decreases the already positive flypaper e ect. The political cost of raising taxation is the reason why an increase in the lump sum grant increases the public good provided more than an increase in private income. The more highly populated the municipality is, the lower the per capita cost of providing the public good becomes, hence the political cost is also lower. This feature can imply a decrease in the flypaper e ect if the initial level of the public good (before the increase in population) is very high, such that the increase in marginal utility (net of marginal disutility due to the political cost of taxation) due to a unit increase in public good is lower than an increase in marginal utility due to unit increase in private On the other is not a > 0 if and only if < 1+ 1+N 2, therefore the public good 7

8 If public good provision before the increase in population is low, the increase in marginal utility due to a unit increase in public good is higher than the increase in marginal utility (net of marginal disutility due to the political cost of taxation) due to a unit increase in private consumption. Hence, the lower cost of providing the public good due to the increase in population implies an increase of its provision and hence of the flypaper e ect. 2.1 Testable Hypotheses We are interested in comparing the change in G (y, t), when an increase in local tax revenue is exogenously induced with the case when an increase in t is introduced. Hence, we can use our theory by assuming that the change in tax revenue on principal dwellings that we observe in our data (which will be described below) is due to an exogenous change in municipalities endowments, which, through the optimization process, (that we described in the previous Section) gives rise to a change in equilibrium taxes a ecting the provided public good. So we test the following Hypotheses: t=0 8 and Hypothesis @N < 0 if and only if > and > 0, which t=0 1+N 2 is the case for expenditure functions financed by the majority of tax Hypothesis @N > 0 if and only if < and > 0, which t=0 1+N 2 is the case for expenditure functions for which the minority of tax revenue is used. 3 Institutional framework Municipalities in Italy are responsible for a wide range of important public programs regarding welfare services, territorial development, local transport, nursery school education, sports and cultural facilities, local police services, as well as most infrastructural spending. Municipalities can rely on two main revenue sources: transfers from upper levels of government (mainly central and regional governments) and own revenues (from own taxes and fees). In what follows we describe the financial feature of Italian municipalities over the years 2006 to 2011, which coincides with the time span of our dataset. The main local tax revenue is a property tax ICI (Imposta comunale sugli immobili) introduced in

9 and applied to real estate. This tax is paid every year by property owners directly to the municipality where the property is located. In particular ICI levied di erently on principal dwellings and on other properties and the tax base is the cadastral income, which does not vary over time. The di erence between the two is the di erent possible tax rates: the maximum threshold is lower for the principal dwellings and deductions are allowed only for principal dwellings. Other important tax revenue sources for the Italian municipalities are the tax on urban waste disposal (Tarsu) which is calculated based on land registry values, the tax on the occupation of public space and a surtax on personal central income tax. Additional own revenues can be raised by Italian municipalities through fees which are linked to the municipal provision of various services. 3.1 The 2008 tax reform Law no. 93/2008 replaced the property tax levied on principal dwellings with a compensating transfer from the central government. As a consequence in 2008 and subsequent years, each municipality received a transfer whose amount was determined by two criteria: a) e ciency in tax collection, given by a1) the ratio between the average value of the revenue of the property tax levied on principal dwellings for the period , measured in cash terms, and a2) the average value of the revenue of the property tax levied on principal dwellings for the period , measured in accrual terms; b) compliance of the domestic stability pact for the year Furthermore, some special exceptions were allowed for small municipalities. Clearly the fulfillment of these two past goals can not be a ected by today s policy maker decisions, making the received per capita transfer for the local policy maker exogenous. Nevertheless, the aggregate amount of compensating transfer received by Italian municipalities in 2008 was about 2.8 billion euro, while the revenue from the property tax on principal dwellings collected in 2007 was around 3.5 billion euro. In order to appreciate the impact of the reform on the composition of the municipal budget we analyze the source of municipal finance for the period (that is the time span we use in the empirical analysis, which will follow). For the period before the reform ( ) property tax accounts, on average, for about 24% of municipalities total revenue: in particular, the property tax levied on principal dwellings is about 8% and that levied on non principal dwellings (buildings, lands, production activities, secondary dwellings) is about 16%. In the same period, current transfers from central 9

10 government constitute on average 19% of the total revenue of Italian municipalities. After the reform (from 2008 to 2011), the total property tax (only applied to nonprincipal dwellings) constitutes about 17% of the total revenue and current transfers from central government are, on average, 26% of total revenue. This increase (from 19% to 26%) in the central transfer quota of the municipal revenue is almost completely driven by the introduction of the compensating transfer which, for the period is, on average, 5% of total municipal revenue. 3.2 Dataset The empirical analysis is based on a dataset for Italian municipalities resulting from a combination of di erent archives publicly available from the Italian Ministry of the Interior, The Italian Ministry of the Economy and the Italian Institute of Statistic. The distinction between revenue from property tax levied on principal dwellings and revenue from property tax levied on non-principal dwellings has only been recorded in Italian municipalities budget since Therefore, our panel dataset covers all Italian municipalities belonging to Regions ruled by ordinary statutes for the period It includes a full range of information organized into three sections: 1) municipal financial data; 2) electoral data covering the results of elections in which the mayors in o ce during the period covered by the dataset were elected; 3) municipal demographic and socio-economic data such as population size, age structure, average income of inhabitants. Since we are interested in testing the flypaper e ect and its relation with the size (population) of the municipality, we exclude from our dataset municipalities that are the capital of the province where they are situated, because their average population (180,000) is by far larger than the average population of all other municipalities (5,500) and this di erence is statistically significant. 5 Moreover, municipalities that are the capital of the province normally provide a much wider range of services than others. Also, we did not include municipalities in regions with special autonomy and other municipalities with missing values from our dataset. Finally we obtain a sample of 5,651 municipalities including 33,906 observations from 2006 to 4 We also collected data for the period since in the analysis which follows we use lags of the dependent variable and of some explanatory variables as instruments. 5 In our dataset the number of municipalities that are the capital of the province is 77 for each year corresponding to 1,36% of the municipalities available in the sample. 10

11 Dependent variable Our dependent variable is the level of per capita current expenditure in each municipality (G), which, according to our theoretical model, we split into two groups: principal expenditure (G p ) and residual expenditure (G r ). The principal expenditure group comprises three expenditure functions, Adminstration & Management, Road & Transport and, Planning & Environment. The total of these latter functions, which altogether are essential in the daily life of a municipality, constitute, on average for the period , almost 70% of the total current expenditure (Table 1). The remaining 30% of total expenditure is for Municipal police, Education, Culture, Sport, Tourism, Social welfare, and also in a very low percentage for Economic development, In-house productive services and Justice. The latter functions are important, but not as essential as the previous ones; in fact many medium-sized and small municipalities do not spend any money on them or they manage these function by networking with other municipalities. ***** insert here TABLE 1 ***** Explanatory variables We build a variable icigrants containing the per capita value of the property tax on principal dwellings from 2006 to 2007 and the per capita value of the grants compensating for the corresponding missing revenue on principal dwellings from 2008 to We then build a matrix of neighbors (W) to each municipality for every year based on geographical contiguity. We then make a row standardization such that the elements of each row add up to one. As a result we have, for each municipality in the period , an average value of its neighboring current per capita expenditure (WG), per capita principal expenditure (WG p ) and per capita residual expenditure (WG r ). We 6 Over 48,606 (8,101 municipalities for 6 years) potential observations, our sample includes 33,906 observations. As a matter of fact, we exclude 8,388 (1,398 municipalities for 6 years) observations referring to municipalities in Special Statute Regions and Province, 462 (77 municipalities for 6 years) observations relative to municipalities that are the capital of the province, and 5,850 observations (974 municipalities for 6 years) relative to municipalities/years where data are not complete or data are missing. 11

12 need this variable since expenditure in neighboring municipalities can be correlated with exogenous controls hence leading to biased and inconsistent estimates of the parameters (Case et al., 1993; Revelli, 2002). As additional variables we include the per capita value of the current grants (netgrants) which are net of compensating grants replacing ICI on principal dwelling from 2008 onwards Control variables We also include a set of time-varying variables which characterize a municipality s demographic, economic and political situation. In relation to demographic control we include the population of the municipality (pop), the population density (density) calculated as the number of citizens per area and the inverse of the population (ipop): these variables can capture the presence of scale economies or diseconomies in the provision of public goods. The proportion of citizens aged between 0 and 5 (child); the proportion aged over 65 (aged) and the proportion of families (families) can account for some specific public needs (e.g., nursery school, nursing homes for the elderly). Regarding economic and financial controls we include the average per capita income proxied by the personal income tax base (income) and the per capita value of the property tax levied on non-principal dwellings (ici2 ). We add some political control that may influence local budget. In particular we set a dummy (election) equal to one for each election year during the period and zero otherwise; we measure the political power of the mayor by using the percentage of votes cast in the first ballot (voteshare). Since Italian law establishes a limit of no more than two consecutive terms of o ce for a mayor, a dummy variable (termlim) has been created to indicate whether a mayor in o ce in a given year is in her second consecutive term of o ce, and thus ineligible for a further term: the impossibility of further reelection may significantly bias the budget-related decisions of a municipality (Besley and Case, 1995; List and Sturm, 2006). The summary statistics, data description and data sources of all the variables used in the analysis are reported in Appendix, Tables A1 and A2. 4 Preliminary evidence As a preliminary piece of evidence it is interesting to look at the mean di erence in expenditure revenue variables before and after the reform (Table 2). In particular, 12

13 average per capita current expenditure (from now on only expenditure ) after the reform is euro higher than that before the reform and this di erence is statistically significant at 1%. The same di erence for both principal and residual expenditure is, respectively, (1% significant) and 4.33 (10% significant). Note also that the per capita revenue from property tax on principal dwellings is, on average, euro and after the reform, the corresponding revenue from compensating grants is euro lower, the di erence being statistically significant (1%). So we find preliminary strong evidence of an increase in expenditure after the reform, even if the available revenue compensating the municipalities was lower. The reform seems to have led to a significant increase in principal expenditures. ***** insert here TABLE 2 ***** We investigate further by focusing on the period , namely the years just before and after the fiscal reform, to test whether there is a di erence in municipal spending behavior according to size. We apply the di erences-in-di erences approach (DD). To do this we use data in 2007 (when the tax on principal dwelling was still in force) and data in 2008 (the first year when tax on principal dwelling was replaced by a compensating transfer). We split the sample into large and small municipalities, where large municipalities are those with a population of over 5,500 inhabitants 8 (the mean) and, small municipalities are those below the mean. We also split the expenditure into principal and residual, as previously defined. In relation to principal expenditure, the di erence in principal expenditure (Table 3- Panel A) for small municipalities before and after the reform (22.12 per capita euros) is larger than the same di erence for large municipalities (16.84) and such di erences are statistically significant at 1%. The di erence of the di erences in principal expenditure between small and large municipalities, before and after the reform, leads to an estimate that is equal to per capita euros (statistically significant at 1%). Therefore, the change in fiscal regime has led to a increase in principal expenditure for both small and large municipalities, however large municipalities increase their principal expenditure less than small municipalities. 7 The restriction to the years reduces the data set to a sample of observations (5,651 municipalities observed twice). 8 Municipalities with a population of over 5,500 account for almost 30% of the sample. 13

14 As it regards residual expenditure (Table 3 - Panel B) we find evidence that the di erence in residual expenditure for large municipalities before and after the reform (17.97 per capita euros, statistically significant at 1%) is higher than the same difference for small municipalities (3.43 per capita euros, statistically significant at 1%). Hence, the di erence of the di erences in residual expenditure between small and large municipalities, before and after the reform, leads to an estimate that is equal to per capita euros (statistically significant ay 1%), implying that the change in fiscal regime has led to an increase in residual expenditure for both small and large municipalities, however large municipalities increase their residual expenditure more than small municipalities. Our analysis suggests that after the change in fiscal regime, large municipalities increased their principal expenditure less than small municipalities (-5.27); on the other hand, large municipalities increased their residual expenditure more than small municipalities (14.54). ***** insert here TABLE 3 ***** 5 Econometric strategy and results Our econometric strategy is based on a dynamic panel data model that also contains a space component. Thus, the dynamic version we estimate (Anselin et al. 2007) is as follows: G it = + G it 1 + W G it + netgrants it + 1 icigrants it + 2 (icigrants it post) + 3 (icigrants it pop it )+ 4 (icigrants it pop it post)+ 5 pop it + 6 (pop it post)+ 0 x it + µ i + t + " it (5) where G it is total expenditure, which we then split into principal expenditure (G pit ) and residual expenditure (G rit ), for municipality i in year t; WG it is the average expenditure of the neighboring municipalities of municipality i in year t, wherew is a matrix of identical exogenous weights (based on geographical contiguity); netgrants it is the per capita value of the current grants which are net of the compensating grants (for the principal dwellings property tax abolished in 2008) ; icigrants it is the per 14

15 capita revenue from the property tax on principal dwellings from 2006 to 2007 and the per capita revenue from grants compensating for the corresponding missing revenue on principal dwellings from 2008 to 2011; post is a dummy variable equal to 1 in the years when the property tax had been replaced by the compensating grant (from 2008 onwards); pop it is the population of municipality i in year t; x it is the vector of explanatory variables described in section 3.2.3; t is a year specific intercept; µ i is an unobserved municipal specific e ect and " it is a mean zero, normally distributed random error. Thus, the coe cient pop it which in Section 2.1, t=0 captures the impact of an increase in tax on principal dwellings for a given level of population and the coe cient in Section 2.1, t>0 captures the impact of an increase in the compensating transfer for a given population level. Our first hypothesis (the flypaper > 0 stated in Section t>0 t=0 2.1, is then verified if pop it 3 pop it > 0, regardless of whether we principal or the residual expenditure as dependent variables. Since t>0 and 3 in Section 2.1, our @N t=0 < 0 is verified, when we use the principal expenditure as the dependent variable, if 4 3 < 0, it means that an increase in population the flypaper e ect. Finally our @N t=0 > 0isverified, when we use the residual expenditure as the dependent variable, if 4 3 > 0, it means that an increase in population increases the flypaper e ect. 5.1 The choice of instruments In order to estimate (5) we use the system GMM dynamic panel estimator (Arellano and Bover, 1995; Blundell and Bond, 1998). This estimator is an augmented version of the di erence GMM (Arellano and Bond, 1991) hence more e cient than the latter (Blundell and Bond, 1998). The system GMM, unlike the di erence GMM, which just employs the di erence equation, builds a stacked dataset, one in levels and one in di erences. Then the di erences equations are instrumented with levels, while the levels equations are instrumented with di erences. The dynamic model we estimate includes the lagged endogenous variable of G it and, in our case, it also includes further endogenous variables: the neighboring spending (WG it ) and the grants net of compensative grants from 2008 (netgrants). These variables are then instrumented by using the other exogenous variables and their lags. In 15

16 relation to the other variables, one might argue about the endogeneity of icigrants and ici2. However, we consider the variable icigrants as exogenous because, on one hand, the tax base of the property tax is given by the cadastral income that is exogenous (for the same reason the variable ici2 is also exogenous); on the other hand, compensating grants were determined for each municipality using previous socio-economic indicators as explained in Section 3.1 therefore must necessarily be perceived by the policy maker as exogenous. The validity of the instruments used in the regression is evaluated according to the Hansen and the AR tests. In particular, in the equation for total expenditure, we start by instrumenting our lagged dependent variable and the other endogenous variables using the standard treatment i.e. using the first order lag to instrument the lagged endogenous variable and the second order lag to instrument the other two endogenous variable WG it and netgrants it. However, it turns out that these instruments are not valid since we reject the null hypothesis of the Hansen test (p-value=0.024). As a consequence we use longer lags, namely the second order lag for the lagged endogenous variable and the third order lag for both WG it and netgrants it. Again in this case we reject the null hypothesis of the Hansen test (p-value=0.029), and we also find second-order serial correlation (p-value=0.078). Finally, using longer lags, we find the combination of lags that allows us to deal with both the serial correlation condition and the validity of instruments. In particular, we instrument the endogenous lagged variable by using its sixth and seventh order lag i.e. using G it 7 and G it 8 for the equations in di erences and G it 6 for the equations in levels; 9 for WG it we use the third, forth and fifth order lag i.e. using WG it 3, WG it 4 and WG it 5 for the equations in di erences and WG it 2 for the equations in levels. 10 Finally, for netgrants it we only use lag 5, namely netgrants it 5 for the equations in di erences and netgrants it 4 for the equations in levels. 11 In this way we do not reject the null hypothesis of no secondorder serial correlation (p-value = 0.523) and we do not reject the null hypothesis of the Hansen test (p-value = 0.354). We also test the validity of any subset of instruments, namely instruments for the level equations, instruments for the lagged endogenous variables G it 1, instruments for WG it and instruments for netgrants it,usingthectest and also in this case, for each subset, we do not reject the null hypothesis that the 9 An addition instrument G it 7 is available but it would be mathematically redundant in system GMM, which is why it is dropped (Roodman, 2009). 10 see footnote see footnote 9. 16

17 specified variables are proper instruments. 12 In relation to the equation for principal expenditure (G pit ), we start again by instrumenting our lagged dependent variable and the other endogenous variables using the first order lag to instrument the lagged endogenous variable and the second order lag to instrument both the other two endogenous variables WG pit and netgrants it. However, it turns out that our instruments are not valid since we reject the null hypothesis of the Hansen test (p-value=0.002). As a consequence we use longer lags, namely the second order lag for G pit 1 and the third order lag for both WG pit and netgrants it. In this case, we do not reject the null hypothesis of the Hansen Test (p-value=0.178) and also we do not reject the null hypothesis of no second-order serial autocorrelation (p-value=0.329). However, by looking at the C-test, we reject the hypothesis of exogeneity for the instruments of G pit 1, namely the instruments are not exogenous (p-value=0.087). Again, we use longer lags and we come up with the combination of lags that allows the tests to be passed. In particular, we instrument G pit 1 by using its fifth and its sixth lag, i.e. using G pit 6 and G pit 7 for the equations in di erences and G pit 5 for the equations in levels; 13 for WG pit we use lags 3 and 4, namely WG pit 3 and WG pit 4 for the equations in di erences and WG pit 2 for the equations in levels. 14 For netgrants it we only use lag 4, that is to use netgrants it 4 for the equations in di erences and netgrants it 3 for the equations in levels 15. In this way we do not reject the null hypothesis of no second-order serial correlation (p-value = 0.777) and do not reject the null hypothesis of the Hansen test (p-value = 0.430). We then test the validity of any subset of instrument by using the C-test and, for each subset, we do not reject the null hypothesis that the specified variables are proper instruments. 16 Finally, for residual expenditure (G rit ) we use the standard instrumenting treatment i.e. the first order lag to instrument the lagged endogenous variable (namely we use G rit 2 as an instrument for the equations in di erences and G rit 1 for the equa- 12 P-value instruments for level equation is 0.376; P-value instruments for G it 1 is 0.289; P-value instruments for WG it is and P-value instruments for netgrants it is The null hypothesis is that specified variables are exogenous. 13 see footnote see footnote see footnote P-value instruments for level equation is 0.190; P-value instruments for G pit 1 is 0.371; P-value instruments for WG pit is and P-value instruments for netgrants it is The null hypothesis is that specified variables are exogenous. 17

18 tions in levels 17 ), the second order lag to instrument the endogenous variable WG rit (we use WG rit 2 as an instrument for the equations in di erences and G rit 1 for the equations in levels 18 ) and the second order lag to instrument the other endogenous variable netgrants it (we use netgrants it 2 as an instrument for the equations in di erences and netgrants it 1 for the equations in levels 19 ). It turns out that the instruments are valid since we do not reject either the null hypothesis of the Hansen Test (p-value=0.307), or the null hypothesis of no second-order serial autocorrelation (p-value=0.868). We also test the validity of any subset of instrument by using the C-test and again in this case, for each subset, we do not reject the null hypothesis that the specified variables are proper instruments Results We do our estimations using the SYS-GMM (Table 4 col. 3 and Table 5, col. 3 and col. 6), which in our framework is necessary to correct the bias and inconsistency of the estimates we would get by using the OLS (Table 4, col.1 and Table 5 col. 1 and col. 4) or, the FE estimator (Table 4, col.2 and Table 5, col. 2 and col. 5). We start considering total expenditure as the dependent variable (Table 4, col. 3). The coe cient of the lagged dependent variable (0.5525) is positive and statistically significant at 10% implying that the total expenditure has a certain degree of inertia. In relation to neighboring expenditure, the estimated coe cient is and significant at 10%, meaning that municipalities tend to increase their own current spending as a response to an increase in expenditure of their neighboring municipalities. The coe cient accounting for the flypaper e ect, 2+ 4 pop it 3 pop it, is positive and statistically significant for any level of population from 13,000 inhabitants, thus confirming the presence of the flypaper e ect (Hypothesis 1). In order to appreciate this e ect consider, as an example, a municipality with population of 13,000 inhabitants, then the impact on expenditure of a unit increase in revenue from compensating grant is given by [ ( ) ( ) = ] which is statistically significant at 10% see footnote see footnote see footnote P-value instruments for level equation is 0.307; P-value instruments for G rit 1 is 0.166; P-value instruments for WG rit is and P-value instruments for netgrants it is The null hypothesis is that specified variables are exogenous. 21 In what follows, all the linear combinations have been computed dividing the population by

19 ***** insert here TABLE 4 ***** When we consider principal expenditure as the dependent variable (Table 5 - col. 3) we find a degree of inertia of expenditure (the coe cient of the lagged dependent variable is and statistically significant at 1%), while we do not find any evidence of horizontal spill-over since the coe cient of the neighboring expenditure (0.0985) is not statistically di erent from zero. The coe cient accounting for the flypaper e ect, 2+ 4 pop it 3 pop it, is always positive and statistically significant as long as the population is less than 15,000 inhabitants, hence confirming the presence of the flypaper e ect for this group (Hypothesis 1). As an example, take a municipality with an average population level (5,500 inhabitants), then the impact on principal expenditure of a unit increase in the compensating grant is given by [ ( ) ( ) = ] an estimation that is statistically significant at 1%. Notice that, the population threshold of inhabitants after which the flypaper e ect does not hold, anticipates to a certain extent the test of Hypothesis 2, which states that the flypaper e ect is negatively linked with the population. However, in order to test Hypothesis 2, we need to compare both coe cients 4 and 3 (see the last paragraph of Section 5). The former coe cient is negative and equals (statistically significant at 1%), the latter one is and statistically significant at 5%. The di erence between the two coe cients is negative [ = ( )] and statistically significant at 10%, implying that an increase in population leads to a decrease in the extent of the flypaper e ect for this group of expenditures hence confirming Hypothesis 2. Finally, when we use the residual expenditure as the dependent variable (Table 5 - col. 6) we again find a degree of inertia in the expenditure (the coe cient of the lagged dependent variable is and statistically significant at 1%) and no evidence of horizontal spill-over (the coe cient of the neighboring expenditure is but not statistically significant from zero). The coe cient accounting for the flypaper e ect, 2+ 4 pop it 3 pop it, is always positive and statistically significant for any given population level confirming Hypothesis 1. Let us consider again, as an example, a municipality with an average population level (5,500 inhabitants), then the impact on residual expenditure of a unit increase in the compensating grant is given by [ ( ) ( ) = ] since in the regressions the variable pop has been rescaled dividing it by

20 an estimation that is statistically significant at 1%. Furthermore, in order to test Hypothesis 3, we compare coe cients 4 and 3. The former coe cient is positive and equal to (statistically significant at 1%), the latter is and not statistically significant. The di erence between the two coe cients is positive [ = ( )] and statistically significant at 1%, implying that an increase in population leads to a increase in the extent of the flypaper e ect (Hypothesis 3). Note that in this case, as we would expect, the flypaper e ect holds for any population level since the relationship between flypaper and population is positive. ***** insert here TABLE 5 ***** 6 Conclusion In this study we investigated the impact on expenditure of tax on principal dwellings before 2008 and the impact on expenditure of the grant which, after 2008, compensated for the abolition of the tax on principal dwellings. This is an interesting reform which allows the existence of the flypaper e ect in the spending behavior of Italian municipalities to be tested. First, we setup a theoretical model in which the introduction of a political bias against taxation gives rise to the flypaper e ect. If the public good is very important with respect to private consumption then an increase in the municipal size implies a decrease in the extent of the flypaper e ect; the opposite happens if the public good is not important with respect to private consumption. The increase in size of the municipality makes the public good cost less and this feature, when the public good is very important, increases the sensitivity of the public good to the grant less than the sensitivity of the public good to the tax. On the other hand, when the public good is less important, the increase in the size of the municipality increases the sensitivity of the public good to the grant more than the sensitivity of the public good to the tax. We then tested the hypotheses coming from the model by using data on Italian municipalities, focusing on two groups of expenditures: the principal expenditure, which should be that essential to guarantee the minimum standard daily life of a municipality and the rest, defined as residual expenditure. We find that the flypaper e ect holds for both kinds of expenditure, but decreases with respect to population in the case of 20

21 principal expenditure and increases with respect to population in the case of residual expenditure. 21

22 References M. Arellano and S. Bond. Some tests of specification for panel data: Monte carlo evidence and an application to employment equations. Review of Economic Studies, 58(2):277 97, M. Arellano and O. Bover. Another look at the instrumental variable estimation of error-components models. Journal of Econometrics, 68(1):29 51, E. Becker. The illusion of fiscal illusion: Unsticking the flypaper e ect. Public Choice, 86(1-2):85 102, G. S. Becker and C. B. Mulligan. Deadweight costs and the size of government. Journal of Law and Economics, 46(2): , F. Blanco. PhD thesis, PUC, Rio de Janeiro, R. Blundell and S. Bond. Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics, 87(1): , ISSN M. Bordignon and S. Piazza. Who do you blame in local finance? an analysis of municipal financing in italy. CESifo Working Paper Series 3100, CESifo Group Munich, D. F. Bradford and W. E. Oates. The analysis of revenue sharing in a new approach to collective fiscal decisions. The Quarterly Journal of Economics, 85(3): , T. Buettner and D.E. Wildasin. The dynamics of municipal fiscal adjustment. Journal of Public Economics, 90(6-7): , A. C. Case, H. S. Rosen, and J. R. Hines. Budget spillovers and fiscal policy interdependence: Evidence from the states. Journal of Public Economics, 52(3): , B. Dahlby. The marginal cost of public funds and the flypaper e ect. International Tax and Public Finance, 18(3): , S. Gamkhar and A. Shaw. Intergovernmental fiscal transfers. Principle and practice, chapter The impact of intergovernmental fiscal transfers: a synthesis of the conceptual and empirical literature, pages World Bank,

23 E. Gennari and G. Messina. How sticky are local expenditures in italy? assessing the relevance of the flypaper e ect through municipal data. International Tax and Public Finance, 21(2): , E. M. Gramlich. State and local governments and their budget constraint. International Economic Review, 10(2): , E. M. Gramlich. Intergovernmental grants: a review of the empirical literature, pages Lexington, MA, J. Hamilton. The flypaper e ect and the deadweight loss from taxation. Journal of Urban Economics, 19(2): , J. M. Henderson. Local government expenditures: A social welfare analysis. The Review of Economics and Statistics, 50(2): , J. R. Hines and Richard R.H. Thaler. Anomalies: The flypaper e ect. The Journal of Economic Perspectives, 9(4): , R.P. Inman. flypaper e ect. In S.N.Durlauf and L.E. Blume, editors, The New Palgrave Dictionary of Economics. Palgrave Macmillan, A. Kalb. The impact of intergovernmental grants on cost e ciency: theory and evidence from german municipalities. Economic analysis and policy, 40(1):23 48, B. Knight. Endogenous federal grants and crowd-out of state government spending: Theory and evidence from the federal highway aid program. American Economic Review, 92(1):71 92, R. Levaggi and R. Zanola. Flypaper e ect and sluggishness: Evidence from regional health expenditure in italy. International Tax and Public Finance, 10(5): , J. A. List and D.M. Sturm. How elections matter: Theory and evidence from environmental policy. The Quarterly Journal of Economics, 121(4): , S.B. Megdal. The flypaper e ect revisited: An econometric explanation. The Review of Economics and Statistics, 69(2): , F. Revelli. Testing the taxmimicking versus expenditure spill-over hypotheses using english data. Applied Economics, 34(14): ,